June 13, 2026 — It's been the most chaotic week in AI since... well, last week. And at the center of it all? Anthropic's Fable 5, a secret sabotage scandal, an unprecedented US government export ban, and China's Z.ai quietly dropping GLM 5.2 on the exact same day. Let's break it all down.
On Tuesday, June 9, 2026, Anthropic released Claude Fable 5 — the first public version of its ultra-powerful Mythos-class model. Fable 5 was positioned as "Mythos-level capability, made safe for general use." Anthropic's blog post described it as "state-of-the-art on nearly all tested benchmarks," with exceptional performance in software engineering, knowledge work, vision, and scientific research.
The pricing was set at $10 per million input tokens and $50 per million output tokens — aggressive for a frontier model. It was bundled into Pro, Max, Team, and Enterprise plans at no extra cost through June 22.
Key capabilities:
But Fable 5 also shipped with very public guardrails around cybersecurity and biology — topics that would be silently rerouted to the weaker Opus 4.8 model.
Sources:
Here's where things got ugly.
Buried in Fable 5's 319-page system card, Anthropic disclosed something unusual. For most safety categories (cybersecurity, biology, chemistry), Fable 5 would visibly fall back to Opus 4.8 — the user would be told. But for one category — frontier LLM development — the safeguard was invisible.
The system card explicitly stated:
"Unlike our interventions for cybersecurity, biology and chemistry, and distillation attempts, these safeguards will not be visible to the user. Fable 5 will not fall back to a different model. Instead, the safeguards will limit effectiveness through methods such as prompt modification, steering vectors, or parameter-efficient fine-tuning (PEFT)."
In plain English: if you were an AI researcher using Fable 5 to build or improve another AI model, Anthropic would silently poison your outputs without telling you. Your model would quietly generate worse results, and you'd never know why.
The AI research community erupted. Ethan Caballero tweeted: "the claude fable 5 nerf for AI research has induced the angriest reaction from AI researchers that I've ever seen in my life."
Dean Ball, a senior fellow at the Foundation for American Innovation, called it "secret sabotage" — writing that "degrading performance on ML research without telling the user is shockingly hostile and a terrible look."
Will Brown, research lead at Prime Intellect, told WIRED: "It felt like Anthropic was saying to the public, 'We don't trust anybody else to do AI research. We are the only ones who have to do AI research.' It feels a bit like they're starting to pull the ladder up behind them."
By Thursday, June 11 — just two days after launch — Anthropic reversed course. The company told WIRED, Fortune, and Business Insider:
"We're changing Fable 5's safeguards for frontier LLM development to make them visible. We made the wrong trade-off and we apologize for not getting the balance right."
Flagged requests would now visibly fall back to Opus 4.8, and API users would receive a reason for refusal.
Sources:
Just when you thought the week couldn't get more dramatic, Friday, June 12 arrived.
At 5:21 PM ET, Anthropic received an export control directive from the US government. The order? Suspend all access to Fable 5 and Mythos 5 by any foreign national — whether inside or outside the United States, including foreign national Anthropic employees.
The Commerce Department cited national security authorities. The letter did not provide specific details about the concern.
Anthropic had no choice. The company abruptly disabled both models for all customers.
Anthropic's statement: "We believe this is a misunderstanding and are working to restore access as soon as possible."
The impact was staggering:
Anthropic argued that the same jailbreak could also be used on OpenAI's GPT-5.5, which is not subject to similar restrictions. The company wrote: "We disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people."
Dean Ball summed up the irony: "An administration whose posture is that we should export advanced AI chips to China, which also wants to ban… Britain (and every other non-American on Earth)… from using our best models? I have no words."
Sources:
Ben Thompson of Stratechery, who has been tracking Anthropic closely, wrote that he wasn't surprised by the secret guardrail fiasco. In his Friday roundup "Hey Siri, Tell Me a Fable," he explained:
"Anthropic released a public version of its Mythos model on Tuesday dubbed Fable 5, complete with a set of very visible guardrails on cybersecurity and biology topics, and silent nerfing around LLM creation capabilities. The latter decision was reversed on Thursday after public outcry, but I wasn't surprised: I explained on Sharp Tech why this sort of behavior was predictable from Anthropic — indeed, it's exactly why I criticized the company in its standoff with the U.S. government."
Thompson's core argument: Anthropic's fusion of belief and business — their genuine commitment to AI safety combined with their $965 billion market position — creates a company that feels "unbeatable" but also prone to these overreaching moves. The same safety-first ethos that makes them build guardrails also makes them think they know best when to apply them invisibly.
This also contextualizes the government's actions. Anthropic had already sued the Trump administration after the Pentagon labeled them a "supply chain risk" in March 2026 — a designation typically reserved for foreign adversaries. The Pentagon wanted the right to use Anthropic's models "for any lawful purpose," including autonomous weapons and mass surveillance. Anthropic refused. The relationship never recovered.
Peter Girnus, a cybersecurity researcher, put it bluntly: "If you describe your product as a munition in every press release, eventually a government takes you at your word. They wrote the legal predicate themselves and called it a brand."
Sources:
And here's where the timing gets poetic.
On Saturday, June 13, 2026 — the same day Anthropic's Fable 5 went dark — Z.ai (Zhipu AI) quietly released GLM 5.2.
While Anthropic was grappling with export controls that locked out every foreign national on the planet, Z.ai's model shipped with:
The contrast couldn't be sharper:
| Anthropic Fable 5 | Z.ai GLM 5.2 | |
|---|---|---|
| Availability | Disabled (export ban) | Live, open to all |
| Context | ~200K tokens | 1M tokens |
| Open source | Closed | MIT license |
| Pricing | $10/$50 per M tokens | ~$18/mo subscription |
| International | Blocked for foreign nationals | Available globally |
| Release date | June 9 (disabled June 12) | June 13 |
Z.ai's timing may be coincidental, but the narrative writes itself: while the US government slams the door on one of America's most advanced AI models to the entire world (including allies), China's flagship model is going MIT open source, accessible to anyone on Earth.
No benchmarks were published alongside GLM 5.2 — a point of criticism from developers. But its predecessor, GLM 5.1, already posted a 58.4 on SWE-bench Pro (edging past Claude Opus 4.6's 57.3) and a 1530 Elo on Code Arena (third in the world). The trajectory is clear.
Sources:
This week in AI history will be remembered for one thing: the collision of safety, geopolitics, and open-source inevitability.
Anthropic — the safety-first lab — tried to secretly sabotage AI researchers using its product, got caught, and apologized. Then the US government — under an administration that has repeatedly clashed with Anthropic — slapped an export ban so broad it locked out allies and the company's own employees. And on the same day, China's open-weight GLM 5.2 launched with a million-token context window and an MIT license.
The AI race isn't just about benchmarks anymore. It's about who gets access, who controls the weights, and whether Western alliances can even agree on who's allowed to build the future.
Anthropic, for all its brilliance, built a $965 billion company on safety. But this week showed what happens when safety shades into control, when "protecting" becomes "locking down," and when the government you warned about AI danger decides you're the one that needs to be contained.
Meanwhile, the open-weight models keep coming.
